作者: Han, C (Han, Chang); Zhang, HY (Zhang, Hongyan); Gao, CX (Gao, Changxin); Jiang, C (Jiang, Cheng); Sang, N (Sang, Nong); Zhang, LP (Zhang, Liangpei)
|
摘要: This paper addresses the remote sensing image fusion problem from the perspective of the analysis sparse model. As an alternative to the synthesis sparse representation approach, the analysis sparse model can yield richer feature representations and better results for image restoration. We, therefore, propose an image fusion method for remote sensing images based on the analysis sparse model. In this method, the analysis operators for the high-resolution multispectral (HR MS) image are trained band by band, directly from the source images, which can greatly improve the adaptability. During the analysis operator learning stage, the geometric analysis operator learning (GOAL) algorithm is utilized with the upsampled low-resolution MS (LR MS) image and the HR panchromatic (HR PAN) image, which does not require an external HR MS image data set. Moreover, the imagery system modulation transfer function (MTF) is considered during the LR MS imaging modeling process, which greatly extends the practical application potential of the proposed method. The simulated and real-data experimental results on IKONOS and QuickBird data sets show that the proposed method can effectively preserve the spectral information and the spatial detail of the image. The fused HR MS images produced by the proposed method are comparable and even superior to the images fused by the other state-of-the-art methods.
|
